Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userModeling Function-Valued Processes with Nonseparable and/or Nonstationary Covariance Structure
106
0
0.0
(
0
)
Added by
Evandro Konzen
Publication date
2019
fields
Mathematical Statistics
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
We discuss a general Bayesian framework on modeling multidimensional function-valued processes by using a Gaussian process or a heavy-tailed process as a prior, enabling us to handle nonseparable and/or nonstationary covariance structure. The nonstationarity is introduced by a convolution-based approach through a varying anisotropy matrix, whose parameters vary along the input space and are estimated via a local empirical Bayesian method. For the varying matrix, we propose to use a spherical parametrization, leading to unconstrained and interpretable parameters. The unconstrained nature allows the parameters to be modeled as a nonparametric function of time, spatial location or other covariates. The interpretation of the parameters is based on closed-form expressions, providing valuable insights into nonseparable covariance structures. Furthermore, to extract important information in data with complex covariance structure, the Bayesian framework can decompose the function-valued processes using the eigenvalues and eigensurfaces calculated from the estimated covariance structure. The results are demonstrated by simulation studies and by an application to wind intensity data. Supplementary materials for this article are available online.
rate research
Read More
In spatial statistics, it is often assumed that the spatial field of interest is stationary and its covariance has a simple parametric form, but these assumptions are not appropriate in many applications. Given replicate observations of a Gaussian spatial field, we propose nonstationary and nonparametric Bayesian inference on the spatial dependence. Instead of estimating the quadratic (in the number of spatial locations) entries of the covariance matrix, the idea is to infer a near-linear number of nonzero entries in a sparse Cholesky factor of the precision matrix. Our prior assumptions are motivated by recent results on the exponential decay of the entries of this Cholesky factor for Matern-type covariances under a specific ordering scheme. Our methods are highly scalable and parallelizable. We conduct numerical comparisons and apply our methodology to climate-model output, enabling statistical emulation of an expensive physical model.
Incorporating covariate information into functional data analysis methods can substantially improve modeling and prediction performance. However, many functional data analysis methods do not make use of covariate or supervision information, and those that do often have high computational cost or assume that only the scores are related to covariates, an assumption that is usually violated in practice. In this article, we propose a functional data analysis framework that relates both the mean and covariance function to covariate information. To facilitate modeling and ensure the covariance function is positive semi-definite, we represent it using splines and design a map from Euclidean space to the symmetric positive semi-definite matrix manifold. Our model is combined with a roughness penalty to encourage smoothness of the estimated functions in both the temporal and covariate domains. We also develop an efficient method for fast evaluation of the objective and gradient functions. Cross-validation is used to choose the tuning parameters. We demonstrate the advantages of our approach through a simulation study and an astronomical data analysis.
Modeling correlation (and covariance) matrices can be challenging due to the positive-definiteness constraint and potential high-dimensionality. Our approach is to decompose the covariance matrix into the correlation and variance matrices and propose a novel Bayesian framework based on modeling the correlations as products of unit vectors. By specifying a wide range of distributions on a sphere (e.g. the squared-Dirichlet distribution), the proposed approach induces flexible prior distributions for covariance matrices (that go beyond the commonly used inverse-Wishart prior). For modeling real-life spatio-temporal processes with complex dependence structures, we extend our method to dynamic cases and introduce unit-vector Gaussian process priors in order to capture the evolution of correlation among components of a multivariate time series. To handle the intractability of the resulting posterior, we introduce the adaptive $Delta$-Spherical Hamiltonian Monte Carlo. We demonstrate the validity and flexibility of our proposed framework in a simulation study of periodic processes and an analysis of rats local field potential activity in a complex sequence memory task.
Two algorithms are proposed to simulate space-time Gaussian random fields with a covariance function belonging to an extended Gneiting class, the definition of which depends on a completely monotone function associated with the spatial structure and a conditionally negative definite function associated with the temporal structure. In both cases, the simulated random field is constructed as a weighted sum of cosine waves, with a Gaussian spatial frequency vector and a uniform phase. The difference lies in the way to handle the temporal component. The first algorithm relies on a spectral decomposition in order to simulate a temporal frequency conditional upon the spatial one, while in the second algorithm the temporal frequency is replaced by an intrinsic random field whose variogram is proportional to the conditionally negative definite function associated with the temporal structure. Both algorithms are scalable as their computational cost is proportional to the number of space-time locations, which may be unevenly spaced in space and/or in time. They are illustrated and validated through synthetic examples.
Mexico City tracks ground-level ozone levels to assess compliance with national ambient air quality standards and to prevent environmental health emergencies. Ozone levels show distinct daily patterns, within the city, and over the course of the year. To model these data, we use covariance models over space, circular time, and linear time. We review existing models and develop new classes of nonseparable covariance models of this type, models appropriate for quasi-periodic data collected at many locations. With these covariance models, we use nearest-neighbor Gaussian processes to predict hourly ozone levels at unobserved locations in April and May, the peak ozone season, to infer compliance to Mexican air quality standards and to estimate respiratory health risk associated with ozone. Predicted compliance with air quality standards and estimated respiratory health risk vary greatly over space and time. In some regions, we predict exceedance of national standards for more than a third of the hours in April and May. On many days, we predict that nearly all of Mexico City exceeds nationally legislated ozone thresholds at least once. In peak regions, we estimate respiratory risk for ozone to be 55% higher on average than the annual average risk and as much at 170% higher on some days.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
